Field of the Invention
The present invention relates to an image generation process performed at arbitrary viewpoint positions through a 3-dimensional reconstruction technology.
Description of the Related Art
In the related art, as 3-dimensional reconstruction technologies, there are technologies for reconstructing 3-dimensional data from images at a plurality of viewpoints at the time of photographing and subsequently generating images similar to images photographed at different viewpoints based on the 3-dimensional data. For example, viewpoints set after imaging (hereinafter referred to as arbitrary viewpoints) can be selected arbitrarily by users. In the technology disclosed in Japanese Patent Laid-Open No. 2009-212728, feature amounts between images chronologically close to target images are calculated when images are generated at arbitrary viewpoints from images at a plurality of viewpoints. When reliability of the feature amounts is low, 3-dimensional data with high accuracy can be generated by reconstructing 3-dimensional data based on chronologically previous and subsequent images.
There is a possibility of regions with no correlation between images at a plurality of viewpoints in an image generation process at arbitrary viewpoint positions being regions that are hidden due to changes in viewpoints (regions shielded by subjects), that is, occlusion regions. For this reason, when 3-dimensional data is reconstructed from chronologically previous and subsequent image information and change amounts of occlusion regions are small between images, there are limits improving mapping precision. Images are assumed to be generated at arbitrary viewpoints after a simultaneous process of mapping all of the images to 3-dimensional data at once is performed. In this case, the amount of data processed at once may be considerable.